Network: Coevolution of Web Behaviour and Web Structure Connor - - PowerPoint PPT Presentation

The Web as an Adaptive Network: Coevolution of Web Behaviour and Web Structure

Connor McCabe, Dr. Richard A. Watson, Dr. Jane S. Prichard and Professor Dame Wendy Hall 17th June 2011

Adaptive Networks on the Web

• Adaptive Web Networks is a growing multi-

disciplinary research area at the intersection of Web & Network Science & Complex Systems.

• Combines the study of dynamics „on‟ (behavior)

and „of‟ (structure) complex networks

• Structure (topology) e.g. (small world, scale free,

community structure, dyads, triads)

• Behaviour (state) e.g. (communicating, blogging,

sharing links, pictures, changing opinion)

Web as a Complex Adaptive System

The Web is not just another complex network, it is a self-

• rganising complex adaptive system (CAS). It co-evolves

with Web user behaviour and exhibits emergent complexity.

• Fig. “2 Magics of Web

Science.” Berners- Lee‟s diagram of how some complexity on the Web can emerge.

Research Questions

• Question 1: How does topology affect behaviour and how

does behaviour affect topology, in different Web networks?

• Question 2: What are the implications of adaptive

mechanisms for Web networks?

State-topology Coevolution Cycle

Gross, T. and Sayama, H. 2009. Adaptive Networks. Springer-Verlag: Berlin

Adapted from Gross & Sayama, 2009

Behaviour Affecting Structure

Dynamical linking (DL), or active linking, describes how actors re-wire links to suit their own individual preferences.

• DL is a key feature of adaptive networks
• Unlike static networks, adaptive networks with DL have

been shown to support emergent phenomena at the macro- level (network level).

• Several theories exist for DL in different contexts, and how

it can be applied e.g. (Hebbian Learning, Homophily and social segregation).

Dynamical Linking at Different Timescales

A separation of timescales between DL & structural process effects nodes state, can result in very different state- topology co-evolution. e.g. Opinion Dynamics Model (ZuErbach-Shoenberg & McCabe et. al 2011).

Initial Network Community structure

zu Erbach-Schoenberg, E., C. McCabe, et al. (2011) On the interaction of adaptive timescales on

• networks. Proc. European Conference on Artificial Life, Paris, France.

Moderate DL Moderate topological effects

Dynamical Linking

Initial Network

Fast DL Slow topological effects

Assortative Mixing

Dynamical Linking

Initial Network Consensus Formation

Slow DL Fast topological effects Fast DL Slow topological effects

Assortative Mixing

Structure Affecting Behaviour

How does structure affect behaviour?

• For Web networks, structure can relate to how documents,
• bjects and web users are linked together. (explicit

hyperlinks, or implicit social links based on interactions) – Structure affects information dynamics: how easily items can be browsed; search engine results, and who connects directly to whom.

• Different topologies of Web networks (small world and

random lattice), can impact collective user behaviour (e.g. Centola, 2010).

State-topology Coevolution of the Web

• 1. Information Networks, (e.g. the Web Graph)

Users may add or remove Hyperlinks when they browse interactive Websites State (behavior) Topology (structure) Website structure contains Hyperlinks which affect user browsing behavior

State-topology Coevolution of the Web

• 2. Micro-blogging Social Network (e.g.Twitter)

Users who receive retweeted messages may form direct links to source. State (behavior) Topology (structure) Twitter social network structure determines what messages are propagated directly

Tweet B A C

State-topology Coevolution of the Web

• 2. Micro-blogging Social Network (e.g.Twitter)

Users who receive retweeted messages may form direct links to source. State (behavior) Topology (structure) Twitter social network structure determines what messages are propagated directly

Tweet Retweet B A C

State-topology Coevolution of the Web

• 2. Micro-blogging Social Network (e.g.Twitter)

Users who receive retweeted messages may form direct links to source. State (behavior) Topology (structure) Twitter social network structure determines what messages are propagated directly

Tweet Retweet B A C

State-topology Coevolution of the Web

• 2. Micro-blogging Social Network (e.g.Twitter)

Users who receive retweeted messages may form direct links to source. State (behavior) Topology (structure) Twitter social network structure determines what messages are propagated directly

Tweet Tweet B A C

State-topology Coevolution of the Web

• 3. Collaborative filtering, embedded user-user

collaborative recommendations e.g. Netflix, Amazon.

When a user buys an item, then it creates a link between a product and user. State (behavior) Topology (structure) Topology of links influences behavior by enabling recommendations to users to buy or sample

• ther products

e.g. Amazon‟s „Frequently Bought Together‟ collaborative recommendations

Implications of Adaptive Web Networks

The hallmarks of adaptive networks (Blasius and Gross, 2009) have implications for adaptive networks in Web Science.

• Robust topological self-organization
• Spontaneous emergence of hierarchies and division of

labour, e.g. (distributed optimization behaviour)

• Complex system-level dynamics, e.g.(self re-inforcing

loops).

Blasius, B. and Gross, T. 2009 Dynamic and Topological Interplay in Adaptive Networks . Wiley-VCH Weinheim.

Summary and Conclusions

• Adaptive network theory and methods offer a formal

framework to study Web complexity ( “magics of web science”)

• State affects the structure of Web networks, and reflexively

the structure affects state on adaptive Web Networks.

• Coupled state-topology generates positive feedback loops
• Dynamic linking produces adaptive Web networks
• Process can happen at different timescales, and lead to

different co-evolved state-topology.

References

1. Giddens, A. (1984) The constitution of society. Polity Press, Cambridge. 2. Miller, J. H. and Page, S. E. (2007) Complex Adaptive Systems: An Introduction to Computational Models of Social

• Life. Princeton University Press.

3. Tetlow, P. D. (2007). The Web's Awake: An Introduction to the Field of Web Science and the Concept 4. Centola, D. et al. (2007) , Cultural Drift and the Co-Evolution of Cultural Groups. 2007. Journal of Conflict Resolution, 51, 6, 905-929. 6. Rupert, M., Rattrout, A. and Hassas, S. (2008). The Web from a Complex Adaptive Systems Perspective. J. Comput. Syst.

• Sci. 74, 2, 133-145.

7. Gross, T. & Sayama, H. (2009) Adaptive Networks. Springer- Verlag: Berlin

References

6. Castillo, C. & Davison, B.D.(2010) Adversarial Web Search, Foundations and Trends in Information Retrieval, Now Publishers, Volume 4, Issue 5, p.377-486.

7. Halford, S., Pope, C., and Carr, L., (2010) A Manifesto for Web

• Science. In: Proceedings of the WebSci10: Extending the Frontiers
• f Society On-Line, April 26-27th.

8. Halpin, H., Clark, A., and Wheeler, M. (2010) Towards a Philosophy of the Web: Representation, Enaction, Collective Intelligence. In Proc. of the WebSci10: Extending the Frontiers of Society On-Line, April 26-27th.

9. Complex systems: A survey, M. E. J. Newman, (2011) Am. J. Phys., in press. of Web Life. Wiley-Blackwell. 10. zu Erbach-Schoenberg, E, McCabe C., and Bullock S., (2011) On the interaction of adaptive timescales on networks", Procs. ECAL 2011, (in press)